Erik: Taylor, thank you so much for joining us on the podcast today.
Taylor: Yeah, thank you very much for having me.
Erik: So, I'm looking forward to this now, even more so because you mentioned that you're a fan of the podcast. That strokes my ego. So, I'm really, really happy to have you on. Also, I think it's a very interesting topic here, a very timely topic. Because we're going to be discussing embedded computing. Because of the chip shortage, this is something that's front of mind for a lot of companies around the world right now. Things that maybe have been hidden to senior management, in some cases, are now front and center. How do we get product out the door when we don't have the perfect supply chain?
Taylor: Yeah, absolutely. It's a conversation we get pulled into a lot these days.
Erik: I'm sure. So, let's get into that topic later. First, I'd love to just understand a bit more about yourself. It looks like you had a passion in this right out the gate. Just based on your CV, you were doing a junior firmware engineer with experimental cosmology lab while you were still an undergrad. Is that right? When did you first touch the topic of embedded computing?
Taylor: Yeah, I think that's a fair place to start. I used to work at the Experimental Cosmology Lab at UBC in Vancouver, where we were building — I was writing drivers in for Linux for, basically, PCI card that was reading out bolometer array so that you could map the cosmic microwave background radiation.
Erik: Let's say, just tracking your career here, you were team lead. I think this is probably also while you were in University for a NASA project, the Regolith Excavation Challenge. Then if I'm correct, your first job was then as a mechanical engineer with Boreal Genomics, where you were — it looks like you were developing prototypes. Is this gene editing? What are you working on here?
Taylor: After working on that NASA Centennial Challenge team, I landed a job at Boreal Genomics. That was a startup, product director of the engineering physics program at UBC. It had a metagenomics application. So, we were building instrumentation for, basically, purifying very long strands of DNA based on interesting technology called 'synchronous coefficient of drag alteration.' But I'll spare you the details maybe. Then from that, we'd leverage that same technology to do a very novel type of, basically, DNA enrichment, a non-PCR-based DNA enrichment. So, there was a lot of complicated instrumentation and really embedded system design related to that, with the goal being how do you extract fragments of carcinogenic DNA from blood so you can diagnose cancer at a much earlier stage? The technology all worked, and the company ended up getting acquired by Natera fairly recently. I think maybe three years ago. But there were some other competitive issues, I guess, you could say. It's a long conversation. It's not really about industrial IoT, so maybe I'll spare you the details.
Erik: Well, that's certainly been a booming area of the market. So, I can imagine that there was a lot of complexity on the competitive side. Let's see. Back in 2014, it looks like you — together with a group of other people — set up MistyWest. Here on your CV, you were first engineer and physicist. Then you've now become the CEO. But you were also a founding employee. So, what did that look like? Did you all know each other, or did somebody set up the company and then recruit you shortly thereafter? What was the background story for the setup of MistyWest?
Taylor: I just, I live in a tight knit community. The company was founded by two other engineering physics grads, basically from the same program. So, I got introduced to them actually working at my previous job. Then myself and a couple other people, we came together. I was one of the co-founding employees back in — this would be like 2014, 2013. We were around six people then. We've just been growing steadily from there working on all kinds of IoT projects, mostly in the commercial industrial space.
Erik: Some really interesting projects. We'll have to deep dive a couple of those. I want to wait a minute to get into the details of MistyWest. But I've got one question. The name MistyWest — I'm from Portland, Oregon, actually. So, I'm imagining this is describing our climate. Is that right? Is this like the misty Northwest? Where's the name coming from?
Taylor: Yeah, I know that. That's where the name comes from. There's actually a bit of a funny story there. So, the company got incorporated quite a few years earlier, I think, actually in 2009 by the two founders, Josh Usher and Leigh Christie. They wanted to create a skimboarding company. Then they shelved that almost right away but picked it back up again after a while, after they cut their teeth working in engineering for a while and decided, hey, they were doing a bunch of consulting work on the side. They were like, "Why don't we just make this a full-time thing?" They were like, "Well, we already have this company rolled out. So, let's just go with that." There's been conversations about changing the name in the past, but it is unique and generic. Sometimes that's actually great, as far as Google is concerned.
Erik: Yeah, absolutely. It's funny because if you look at companies — large and small — there's really no best practice for names, right? Ours is IoT ONE. You can imagine that was a fairly conscious choice to tether ourselves to this topic of IoT. But then, the business expands and you start looking at it and say, "Well, maybe I should choose something a bit more generic, a little bit more amorphous." I don't know. But in the end, people have no problem. We're a consultancy. People ask us to do all sorts of things that have no relation to IoT. So, I think at the end of the day, people are working with the team more than they are with the name.
Taylor: 100%, yeah.
Erik: So, before we get into what MistyWest does in particular, I'd love to discuss a bit more of the topic of embedded computing and how that's changed. I think one good frame to think of that through is this project that you were working on with NASA back in 2009. It sounds like you were working on a hardware for excavation, I guess, on foreign, on the moon or planet. The topic then would be, what has changed between now? Then how would you design differently given the toolkit that you have available now versus the toolkit that you had available then? Then, also, probably equally interesting, what has not changed so much? Are there particular aspects, either of the technology stack or the design process, et cetera, that have remained more or less static over the past decade?
Taylor: That's a great question. So, obviously, that was a long time ago. Some of the things that have changed a lot is — when we were working on that project, Arduino was just getting started. The Arduino Uno was the thing. Now we're seeing companies like Sony, for example, release their Spresense line. They'll talk about how they have Arduino-like interfaces and functionality. So, you have these major Silicon manufacturers who are standardizing around the Arduino IDE, just because so many people know about it. That's a big change, obviously.
We also used LiDAR for the project. Just to give a little context on the project, too. So, this is one of NASA Centennial Challenges. They also had a Space Elevator Challenge. I think they had one around a glove solution for astronauts. These were, I think, launched back in 2006 or 2007 when NASA announced that it wanted to return to the moon. So, the Regolith Excavation Challenge was really about, hey, if we're going to build a permanent moon colony — which is they're supposed to be having people landing on the moon in the next two or three years here — using the dirt that's there. The regolith that's there is really important. So, this was basically a competition to find new ways of moving that dirt around.
We had to design a relatively autonomous robot. So, yes, we use Arduino microcontrollers for some small pieces of it. But we also needed a way to navigate with a time delay, because they were expecting having a human controller back on Earth. There's 15-minute time delay which makes it a little hard to control. We settled on a hook coil LiDAR solution that was basically line scanning and had something like a 15 FPS frame rate. I think it was $5,000 or $6,000 USD at the time. Now you can go and look at the Velodyne Puck or some of the solutions that are asked and get something that is like a 2D sphere with a much higher pixel density and 10 times the refresh rate for that price point. LiDAR has also come along ways.
In terms of what stayed the same, I think there are some things that really haven't changed too much. The actuators, they're pretty similar. A lot of the wiring and electronics’ design is pretty similar. We're still writing firmware occasionally on Arduino depending on what it is. We use a lot of STM32s these days. I don't know. There are some of those core things that haven't changed too much.
Erik: I imagine, let's say, back then, if you were to be operating on the moon and you have a 15-minute latency, you need the equipment to be relatively self-sufficient in terms of taking actions, I suppose. Otherwise, it's going to take you forever to do anything, because you're going to be sending commands, assessing the situation. So, edge computing. What do you see is the potential for edge computing to — maybe how are we doing this when we're on Mars today with a rotor? I guess, to some extent, that's what we're doing, right? There's no real time or even — I don't know what the latency is there. It's probably hours, right? So, how are we managing that today on this mode?
Taylor: I don't know offhand, but I think it is on the order of hours. They do path planning. I think the robot has some ability to, basically, make some decisions on its own. If it gets into trouble or, say, a torque on a certain motor exceeds a certain threshold, or whatever, it will stop and allow the driver to reassess. But it's not super-fast moving around. They're really careful with how they drive these things, because they don't want them to get stuck. In terms of having a fully autonomous vehicle just making decisions on its own, that's not really something that is getting done right now, I would say. Though, obviously, going back to the moon, that's something that's going to be, I think, of increasing interest. The moon is a much easier place to deal with than, say, Mars. A 15-minute delay versus a multi-hour delay is a pretty big advantage.
Erik: Yeah, exactly. So, that's from a technical perspective. In terms of the way that you work, maybe this is just going to reflect to an extent. You as you are today with something like 15 years of experience versus you as yourself back then. But in terms of how you manage an embedded computing project, have the best practices changed? Are we fundamentally walking through the same steps just with a different toolbox?
Taylor: Yeah, I think some of the things have gotten easier. So, there's things like the Yocto project for Linux that allow you to basically use recipes to generate your own version of Linux in a relatively straightforward fashion, so that there's more firmware libraries. There's also a much bigger open-source hardware community now, which can be really great. It also means there's more things to choose from at the module level. So, you can buy time-of-flight depth sensor and things like that. There's lots of different options there if you wanted something, as opposed to having to figure out how to make your own structured light sensor from scratch.
So, there's a lot of things that make some of the early product development easier, but a lot of it has stayed similar. The reality is, with product development, there's always a lot of uncertainty. Honestly, some of the biggest uncertainty we deal with and we help our clients with is around actually validating their business model and market uncertainty, where it's very easy to get people to commit on paper to, "Hey, we want to buy this thing at that price point." But when you actually put the thing in front of them, and it has all of its flaws and all of these issues, and there's edge cases that have been handled, sometimes those things don't really materialize. So, we always argue or advise our clients to go with a minimally viable approach and try to iterate what is the minimum thing that you could sell that solves the problem — the core problem related to your company's mission. How do we get there as quick as possible and try to get out into the market and validate that? Then iterate from there and make improvements, and all of that stuff. That piece has really stayed the same. Though, I think, if you go back 15 years, waterfall was a lot more common. The methodologies were less agile. So, there's practices from software that have been slowly percolating into hardware more and more, which is a good thing.
Erik: Yeah, I was discussing, over a coffee this morning, with a company that is an advisor of a company that's building a mechanical solution for energy storage. They're trying to fundraise now. In order to fundraise, they need to prove that there's customers that have some significant interest in the solution, when the solution is really five years away from generating any revenue. How do you do that? This is a bit of an extreme case where they have a long path to revenue. But I think it's not that extreme. I think with a lot of hardware, there's some multiyear path in lining that up in advance.
Maybe this is a good jump off for what you guys do it at MistyWest. So, this actually surprised me a little bit that you're so involved there. Because your website is very much around the technical capability. Obviously, you have a very deeply technical team for developing solutions. Can you give us the high-level value proposition? Then we can look a little bit, step by step, on where you work with companies along the path.
Taylor: Yeah, absolutely. As you're mentioning that example, in a bit here, I'll mention a company that has that same challenge in that there's a very long horizon with this deep tech to getting it to a place where you can actually make revenue. How do you bridge that is a real challenge. In terms of what we do at Misty West, we're an engineering consultancy who offer product development services. So, we try to pride ourselves in offering full stack services. We have a couple of industrial designers and mechanical engineers, electrical, firmware, and software.
Basically, we want to be a place where you come and you might have something you want to measure. Maybe you need a camera sensor or some novel sensor. You want to measure that and get that data into the cloud so you can do something with it, make some inferences. We'll help build that IoT solution out, basically, as a full stack team. We primarily work in the commercial industrial space. So, if it's consumer good, if you want to make a smart dog collar or something like that, we'll probably refer you to someone else who has a bit more experience at the true high volumes. Then if it's something that is a bit more on the deep tech side, we're a company founded by engineering physicists. We have an engineering physics PhD and staff who used to lead the Queen's University Engineering Physics Department. We have a very deep understanding of first principles, understanding of some of the underlying problems for these harder technology ideas.
Erik: So, I'm looking at the website and some of your examples here. You have a really broad range, which I think really reflects the long tail of solutions in the industrial IOT space. We're really looking at a lot of niches that can be extremely meaningful, but it's very different from consumer and that everybody has — all consumers have a similar set of challenges. Whereas when you're looking at the industrial space, you just have these tens of thousands of niches. But you have everything here from polar bear trackers for the World Wildlife Foundation to outdoor signage solutions for Amazon. Who would you typically be working with? It looks like it's a mixture of almost nonprofits, some startups, and then some very large corporates. How do you balance this quite diverse portfolio?
Taylor: Yeah, it's definitely a mix. So, we've worked with in the past with Amazon, Google, Microsoft, Facebook, all of the large FAANGs companies. Typically, we work with their research groups. We also work with some midsize companies and some startups. We try to keep a healthy blend, to be honest. We don't want to have one monolithic customer. Consultancies like ours that tend to do that can have a bad experience when that company decides to move on.
Who do we work with at those companies? For the FAANGs companies, it'll be someone in their research group. Often Microsoft Research or Google X, they might have some ideas they want to quickly try and find answers to and fed out. You sometimes will have relevant expertise for those questions and allow them to just move really quickly and get some answers. Then for your midsize companies, we can talk to, say, a product manager potentially, or a head of R&D, or somebody, really even VPs, or C-suite who have an initiative of mind. They have a problem, and they need somebody who can execute on an IoT solution, building a custom hardware and basically integrating on the software backend side. Then with startups, I'll often find myself playing a bit of an advisory role, helping them understand how do you go from where you are now to seed funding, to your series A, and what does a product development roadmap look like that is actually feasible, when should you hire staff versus when should you look at hiring a consultancy.
One of the roles I hold right now is as an investigator at CDL, Vancouver — the Creative Destruction Lab. It was spun out of University of Toronto business school. I'm forgetting the name offhand. But they have 20 chapters worldwide or something like that now. That's the role I play there. So, we'll advise them at first. When it makes sense, if they really understand what their problem is and it becomes an execution problem, it's like, "Okay. Let's go fast." That's where we'll step in and help. But really, we're just trying to have an impact.
As a company, we have aligned ourselves to the UN Sustainable Development Goals. So, we're looking for projects that help advance those. So, things related to clean tech, clean energy, those kinds of things, inclusive technology. We don't want technology that really only benefits some narrow niche of society. So, we look for projects like that. The WWF Project is an example, where it's not the most lucrative thing. But global warming is a real concern. Polar bears are at extreme risk. For that project, what they had was a collar dating back to the 1970s. Using a satellite network, there are a bunch of problems with that. Not only is it a risk for the bears, but they fall off all the time. The cost of putting a collar on the bear is something like $50,000, because you need a helicopter and a tranquilizer gun to even go put it on. You can only put it on the female bears. Because the male bears, their necks are so wide. They're wider than their skull. So, we help, basically, take that technology into the 21st century. So, we did a bunch of RF design to get the antenna miniaturized, and then used more modern batteries, more modern electronics, optimized everything for low power and low temperature really. Because this thing needs to survive for around a year on a polar bear's ear. So, that was one of the projects that, I think, is maybe a bit more impactful and definitely more interesting.
Erik: Yeah, interesting. I guess, if you're doing something for Google, you're basically trying to help them validate some probably technical concept. If it's for World Wildlife Foundation, I suppose they already understand the end solution, and they're looking for a technical solution that meets the specs. But they already know the application has value and what that is. If you are working with a startup, then there's this fuzzy area where they have a concept. They have maybe some loose validations, having conversations that point them towards a direction. But until you get the product in somebody's hand, it's really hard to understand what interest in that solution is going to be.
How do you, as somebody that's helping to bridge — that's the valley of death where a lot of startups failed to make it through — how do you manage that tension between the need to get as robust and credible customer feedback as possible, which would push you towards investing heavily in a more comprehensive solution versus the need to manage risk and go quick, which would push you towards having a very lean solution that's just a Barebones prototype? I guess you're iterating back and forth. But what would be the journey? Let's say, somebody comes to you with a strong concept and wants help getting this to a stage where they can go to manufacturing, how would you break that down in terms of development, feedback, development, feedback loops?
Taylor: So, in the case of a startup, I can actually pull an example for this. With one company we're working with, Ideon AI — they're a spin-off from TRIUMF, a local particle accelerator — they're doing muon tomography. They first started talking to us maybe four or five years ago. Muon tomography is basically, if you think of X-rays, muons are like X-rays but much higher energy so they can go through up to a kilometer of rock. Muon tomography, usually, where it happens these days is like the Large Hadron Collider. If they want to understand particle physics, they're looking at muons. But there's actually muons showering down on the earth every moment from high energy particles hitting our atmosphere. So, if you can figure out how to measure those underground, you can basically infer the density of the rock that that muon traveled through, which allows you to make line of sight or direct line measurements of significant amounts of rock.
A good way of understanding the technology would be using an analogy. If you think of what MRI scans have done to enable minimally invasive heart surgery like stents, a technology like this could have a similar impact on mining, which is a really relevant thing nowadays given the need to transition to clean energy. So, there's expected to be something like a $12 trillion shortfall in terms of cobalt, lithium, nickel, and copper in the next, I think, 30 years or something like that. Between now and 2050. So, it's a huge, huge problem. Mining companies don't really have a great answer to exploration right now. It's very hit or miss. A lot of the obvious surface minerals aren't easy to — if there's an ore vein on the ground, by now it's been mined out for the most part. So, they need to start looking at subsurface solutions.
This startup came to us four years ago. Just a background on the application, just so people understand. They came to us four years ago. They built, basically, a box, a 1x1x1 meter box using, basically, the exact technology out of the particle accelerator. They just took that. They took some dev kits, slapped it together, and threw that in a mine somewhere. That's really a great way to go about it if you can. Just do something minimally viable. Try to do it with sweat equity. Try to call up whoever you can. Maybe you have a friend at BHP or whatever, and get them to let you stick your detector in their mine for a little bit to collect some data. Then use that in the size of the problem you're solving to generate some interest. Raise that series A. Look for some non — not series A. Raise seed funding, and look for some non-dilutive funding.
A lot of these startups we work with, especially the deep tech space, it's very important for them to find non-dilutive funding at the start. There's a lot of government-based funding. In the US, you can find funding through DARPA. There's a bunch of other. I think there's an energy group, as well, like DARPA. I can't remember the name of it offhand. Then in Canada, there's a whole bunch of other bodies like NSERC and entities like that. So, finding that non-dilutive funding. Then when you get to a place where you've got that initial data, you've got some people who are excited about it, that's when they came and to talked to us. We, basically, advised them. They were looking at miniaturizing. They wanted to get this 1x1x1 meter box into a 10-centimeter circle so that they could put it in a borehole, so that now you don't need to have a mine existing for you to use this technology. You can just drill a hole and throw it in the ground. That's a pretty big challenge. It's a big engineering lift. They had a good understanding of what the opportunity was there. So, that's a place where it made sense to engage to someone like us.
We basically advised them. We said, "Hey, don't look at that Microsemi PolarFire FPGA. Because even though it has good specs, it has very poor documentation. You're going to have a hell of a time finding a developer later on to support that, because it's a very niche product." We, instead, pushed them towards like a Xilinx, Inc or something where it's an FPGA that has a very good adoption, generally in the community. It has things like a dedicated Linux SOC sitting right next to it which can handle off Ethernet communication, things like that. That kind of high-level advice. We basically took them from that 1x1x1 meter box and got them down into that 10-centimeter circle.
Now they're deploying — I don't know. I think they've deployed something like 50 detectors in the wild. They're looking to deploy 500 in the next year. We're basically helping them scale up all of their manufacturing and things like that. We're slowly basically working ourselves out of a job there as they scale. There's pieces where it really makes sense for us to hold on to things, but there's also places where we'll help them to hire quality engineers and set up their manufacturing and things like that.
Erik: Great. That's a perfect example. I mean, it's such a niche case. At least, they're very clearly who your customers. There's no uncertainty about the question and the customer. But proving out the technology, I'm sure, is a costly effort. In this case, just to understand the solution, now you drill a borehole. Is it you're drilling a horizontally on a slope, and then you're going from the ground, then you have the receiver above ground, and you're looking through? Or can you drill vertical and then have a diagonal? Just want to understand how that would work in practice. I'm just curious about this case.
Taylor: Yeah, you can drill. You can drill at a vertical or at a diagonal. Then the way the technology works is, there's basically these scintillator fibers that are interlaced in a special way inside the detector. When the muon passes through the detector, you can tell by which scintillators it hits and when. So, there's a very precise timing problem. You need to have like hundreds of picosecond accuracy on timing for when photons arrive at the detectors at either end of this. It's basically a three-meter tube with a 10-centimeter diameter, in a 10-centimeter diameter circle. You can use that to know what direction the muon came from. The muons are coming in at all different angles. They just come in randomly. They're not going to come from the other side of the Earth, obviously. Because they're not going to pass through the underside of the earth. But you know if one came through where it came from. From that, you can also measure how much energy it has. From that, you can get an idea for how dense was that rock it came through.
Now, the one challenge here is that there's not a ton of muons that are coming through all of the time. It's not like you just take a picture with sunlight. You have to leave those detectors in the ground for a little while to get enough muons to go through them. But the alternative, really, to get a picture like that is drill 300 boreholes instead of 6. So, it's a pretty compelling value proposition when it cost you upwards. It can easily cost you $50,000 per borehole or something like that. Then when it comes time to do extraction, you have a much better picture of what it is you're looking at potentially. It doesn't work for everything. Because not everything has a density contrast. But it's very interesting when it does.
Erik: Okay. That's great. That's precision mining. Aside from saving money for the mining company, this saves a lot of — every time you drill a borehole, you're probably clearing forest and so forth to get in there. So, there's an environmental element here as well. There was another case that you mentioned. Maybe a bit of a different perspective here. But this is a smart city application, where you're building your own IP. So, this was a question I had. Obviously, you're in the situation where you have a lot of technical competence, and you're constantly being confronted with interesting challenges. So, I've got to imagine you're sitting over lunch every day, and somebody says, "Hey, why don't we do this?" It's always a temptation for an entrepreneur to stay focused. But what are you doing in terms of exploring areas where you can build your own IP today?
Taylor: The chip shortage really led us to look at things and say, hey, our clients can't get what they need anymore. We're used to using solutions from NVIDIA, from NXP, even Raspberry Pis. A lot of those were going early end of life or just not available. So, we were looking around and say what can we do about this. We start talking to Renasas in realizing, hey, their fab situation is actually a lot better than some of these other companies. They work with TSMC. They work with Samsung. They have fab in Japan, as well. They actually have parts that really checked all the boxes. So, we started developing a SOM with them.
Once we started doing that, it's funny how the universe talks to you. We started having conversations with other people who had these problems. Another local company, Novax Industries — who they work on pedestrian safety — we've ended up talking to them. We realized that they have really a whole understanding of this pedestrian tracking problem at intersections that really needed a piece of technology that we were already working on, because of this SOM we were developing with Renesas. We've been working with them, bringing together some of our hardware expertise and some of our expertise in computer vision. We do a ton of computer vision work. They're bringing together, really, a deep understanding of traffic problems, what it means to sell traffic systems to the city. Also, they have their own solutions in terms of automatic traffic control and power systems that are uniquely required to work at traffic systems. So, they exploit the solution. We're exploring there in that case, so that the SOM we're working on is based on the Renesas RZ/V2L. It's an Arm SoC that has a dedicated neural processing unit. Basically, it's a piece of silicon that does special linear algebra operations that are perfect for running object detection in objects and image segmentation algorithms. If you think basic convolutional neural net stuff that you can generate in TensorFlow in a known and next format, this SoC will take that and run it.
It's basically comparable to Jetson Nano, but it uses half the power. It costs a lot less, which allows potentially a much lower cost way to do pedestrian tracking at the edge. You can run the compute at the edge, because this thing is cheaper. So, now we're looking at other solutions. Obviously, everyone — there's a lot of people looking at this problem. There's been a significant uptick in pedestrian fatalities in intersections during the pandemic and even before that. So, now a lot of cities in North America are basically signing on to this Vision Zero pledge where they want to try and get to zero pedestrian fatalities. So, there are a bunch of companies. Bosch, for example, has developed a solution in this space. I think they're selling the camera for like USD$3,000 or $4,000 per camera. But it also comes with a subscription. Hardware as a service. They stream. You basically stream all of your image data to the cloud. It's something like $2,000 a year for the first year and then $500 from then on, which is a pretty expensive model for the city to sign up to.
We're thinking that makes sense, I think, if you're doing the compute in the cloud. But some of these applications, I think, in the long run, as to compute — like the example I just gave, where you basically have something that's two or four times more performant than a Jetson Nano. The Jetson Nano is not an old piece of tech, right? So, there's a real trend here and industry. What we're imagining is going to happen is that some of this is going to make sense to move to the edge, just from an economics perspective. That doesn't mean that there wouldn't necessarily be a hardware as a service model here. But if you can push that compute up to the edge, then potentially you don't necessarily need to use all that bandwidth. You don't need to store a bunch of things. So, we're working with them on that. We're developing support for cameras and things that will be good options for low light and outdoor.
Erik: Yeah, and I would imagine in a case like this, machine vision at a pedestrian crossing, latency is going to be critical here. So, a cloud solution might work in 99% of the cases. There might be a storm or something that disrupts the system to some extent. I think that's also a consideration here. This is an interesting trend, though. I've interviewed over the past, I think, year or so a couple of companies that are more on the design side in the semiconductor value chain, designing chips specialized for applications like machine vision. I think that's one trend of saying, okay, we now have enough scale in some of these applications to do design specifically for them. Then we can be much more efficient in terms of how we design the packaging or how we design the chip itself. That still takes a lot of time to get off to market, right? Because they're going to have the same challenge trying to get this into a fab and get manufactured as anybody's having today.
I think your approach is quite interesting to say, okay, you actually found a situation where you had excess capacity. With Renesas, it sounds at least less pressure than some of the other manufacturers are having. You were able to design that for a specific purpose and basically repurpose that. So, I think that's an interesting approach to navigating the challenges we have today.
Taylor: Yeah, well, we were looking around for something that addresses the issues that our clients are currently having but also is a bit future-forward and looking at the problems that they will be able to solve potentially. It's a brand-new solution. We're benefiting from being on the bleeding edge a little bit there. The other thing to say, too, is it does have a pin compatible version, the RZ/G2L, which is more similar to, say, a Raspberry Pi solution. We're using that in other places. If you want to have an IoT gateway, it's a two for one in that regard. It really was a no-brainer once we looked at it a little bit more.
Erik: How far do you guys extend into machine learning? If we're looking at the muon example, I imagine you have a certain density. But then you could say, over a kilometer, we have an average density. It could be in some cases, could be that it's one specific area. Like 20 meters of that is extremely dense, or maybe there's a moderate density across the entire range. So, I guess there's different configurations that could lead to some average. I assume you have complex models behind that that helps interpret the data. Do you, as Misty West, do you get involved in helping companies to develop the ML, or do you step away at the end of, let's say, the border of the hardware?
Taylor: Yeah, it's an interesting question. Primarily, where we expand effort is when it makes sense to get these things on the edge. We'll invest the effort in terms of firmware development and things like that to get them running well on the edge and get them deployed at scale on the edge. In terms of Ideon, for example, they have their own expertise in-house that is really relevant to the problems they have to solve there. It's actually in terms of turning muon tomography data into meaningful actionable insights for a mining company, there's a lot to be done there and a lot of really domain-specific knowledge.
There's actually other sensor modalities that they want to integrate. So, things that are more industry standard right now like magnetometry and gravimetry. But overlaying those datasets with muon tomography data is really complicated. Because the nature and how they measure are so different. Those are things where like those are real ML/AI problems. There's a real challenging inversion problem to solve there as well. But it's not a place where we have a ton of expertise. It does make sense, I think, that I would advise my clients for things like that often to hold that expertise in-house. Because it is going to be really critical for them in terms of in locking their full value proposition.
Erik: Okay. Got it. So, we've been discussing a bit where you are today. I think this area is very interesting because on the one hand, it's a long tail. We're dealing with B2B environments. Also, for large corporates, it's validating the concept matters a lot. It's very difficult for companies to move forward with a new solution development until they either see one of their internal units succeed, or they see a couple of competitors start to develop a solution that has some traction on the market.
I feel like we're starting to see that in high enough volumes, that companies like — I know, here in China, a number of multinational pharma companies are starting to look into IoT devices for helping people recover from Alzheimer's and other diseases and so forth. We're reaching that critical mass where executives are looking at it and saying, okay, we have confidence. We have confidence that collecting data from IoT devices, analyzing it will allow us to help solve business problems. Then they have to still figure it out the user journey, and the business model, and so forth. So, there's a set of challenges there that the companies are working on. But at least, I feel like we've reached that area where the confidence exists that these solutions in many areas can address real needs.
If we look forward to the next five years — well, first, do you agree? Where do you see that market? Then if we look forward to the next five years, what would be the things that excite you about what is going to be possible, whether it's trends in terms of how companies are, from a business standpoint, how they're investing, their ambition level or whether it's from a technical standpoint in terms of how developments will enable capabilities that previously were very challenging and costly to build? Let's cover this in two parts. First, where do you see us today? Then looking out over the next five years, what excites you about the potential for development here?
Taylor: I think in terms of where IoT is today, it's a really exciting space. It is definitely a bit of a red ocean in some ways. So, if you look at — just taking, for example, IOT frameworks, like solutions in the cloud that integrate IoT devices. I probably come across one new example every two weeks or something like that. Siemens has MindSphere. There's a company called Losant. There's Balena for fleet management. There's, obviously, Azure and AWS IoT. There's Ubuntu Core. There's EdgeX. That's just a few. I have a list that's like another 20. It's long. There is a ton of diversity in terms of what's out there. I think what you're going to see in the next five years is some consolidation in those areas, where we start to see some winners and some losers.
That said, I think, maybe backing up a bit, I think I might get the numbers slightly wrong here. I think back in 2010 or something, Cisco was forecasting what they were expecting to see in terms of IoT growth in the next 10 years. I think they were off by a factor of 50% or something like that by 2020. I think the reason why they missed their estimate is because the reality with IoT, especially in the commercial industrial space, is there's a lot of unique components and different systems. There's a lot of diversity. It's not like the iPhone where everyone is expecting a certain user interface. Everyone is just going to go buy an iPhone, and they'll be happy with it. The reality is that every factory, every store, they all have different interfaces and different needs. There's always going to be some degree of customization there. So, I think there's going to be some consolidation in the next five years. But also, I don't know that it's going to go full hockey stick as a result of that. I would still expect there to be some niche things happening.
Then on the technology front, the example I brought up with is Renesas SoC, that's going to keep happening in the next five years. I think there are applications right now that people aren't quite thinking about that are going to start to become viable, that the people just aren't really expecting. Maybe not in five years, but maybe in 10. You could imagine a situation where you buy a dashcam or something that is battery-powered. You just stick it to your window or whatever. You leave it there, and it has its own machine learning algorithms running. Whenever something interesting happens, it'll upload that to the cloud or something like that. That thing could cost like 10 bucks. It's almost like a disposable. That kind of technology, you could see that in the 10, 15, 20-year timeframe. Maybe even earlier.
There's a bunch of technologies that are enabling it. One of them is the compute element that I was talking about. Not only are computers getting even more powerful with these more dedicated processors. But the laptop I'm on right now, it's a Microsoft Surface Book and it's based on ARM. It uses a quarter of the power of its Intel predecessor. So, it can be powered over a USB-C connection. Those are both examples. But also, the battery power density is going up as well. Cost is going down for those batteries. Solar cell efficiency is going up as well. So, at some point, you get to these situations where cost is also dropping. You can have these very ubiquitous, connected devices, and they'll just start to unlock applications that right now just don't really make sense and are a little bit hard to think about.
Erik: I think that's a great point. Often, we end up talking about, to some extent, the cutting edge of it. The topics like edge AI that are a bit more exciting, but a lot of the most meaningful innovation is about simplifying. It's about reducing the cost structure. It's about simplifying deployment. That's what makes it possible to do widespread deployments, to make a business case. Then in the end, to really impact how people live and work in the real world, getting the product out to market.
Good. I think we've covered a bit here. I guess, this is an interesting topic because you could take it from so many different angles. Is there anything critical that we haven't touched on? We haven't even touched on security, and so forth. That's a whole another rabbit hole that we could get into. But anything else that we haven't touched on here, that's really critical to the work that MistyWest is doing?
Taylor: I think we've touched on most of it. We do a decent amount of work on wearables, Bluetooth connected wearables as well. One of the companies we work with — Fatigue Science — they do their own, basically, metric for fatigue. Right now, they're working primarily in mine sites and things like that. But we were originally helping them when they were working with the Cubs. When they won the World Series, they were all wearing wristbands that we'd helped develop. I think that's another space where there's a lot of changes going on, too. When we were working with them, it was maybe five years ago now. But since then, Fitbit has released their own SDK. They got bought by Google. A lot of these wearable technology place have changed from making wearable hardware to software place running on wearable SDKs, where you'll run your algorithm in an app that runs on the Apple Watch or whatever. Really leveraging that hardware, that's definitely one trend we've been seeing. I don't know if there's too much else to mention. Obviously, it's a topic that we could just keep talking about, but maybe I'll park it.
Erik: Yeah, well, maybe that's something to look out for the future. What are the business models of the future when you have — because I also know a lot of companies that are looking and basically saying, we can use some hardware to create a data set and a gateway to the customer. Then we just open that up and let people build things on it. We collect a toll along the way. A lot of room for business model innovation here. I think that's always for us a very exciting place. IoT One, we're not as technical as MistyWest. So, we're much more playing on this side of the equation.
Taylor: It's definitely an important element of it. It's something that, as I get more and more experienced, I learned to appreciate more and more that getting the business model right, figuring out how you're going to monetize is not an insignificant part of the journey often.
Erik: Yeah, exactly. Great. Well, if somebody is interested in having a conversation and learning more about what you guys do, and maybe doing a project with you, what's the best way for them to reach out to the team?
Taylor: There's just our website, mistywest.com. There's a contact form in there. You can reach out to us. You could also reach out to me at firstname.lastname@example.org.
Erik: Awesome. Taylor, thanks for taking the time. I really appreciate it.
Taylor: Yeah, thank you so much.